History and Pastes

Claude Code maintains a global command history and a content-addressed paste cache to enable quick recall of recent prompts and large pasted content.

Command history

A global index at ~/.claude/history.jsonl records recent prompts (JSONL format—one JSON object per line).

Structure

Each history entry contains:

{
  "projectName": "my-project",
  "timestamp": 1713374230000,
  "prompt": "Show me the latest API changes",
  "pastedContent": [
    { "id": 1, "type": "text", "contentHash": "abc123def..." },
    { "id": 2, "type": "image", "contentHash": null }
  ]
}

Indexing and retrieval

• Indexed by project name: Entries are tagged with projectName to separate history by project context
• Newest-first within each project: Entries for a given project are sorted most-recent-first
• Cap: 100 entries per project: Older entries are dropped when the limit is exceeded
• Access: Via prompt up-arrow (↑) in the CLI; cycle through recent prompts for the current project

Large paste handling

Pasted text longer than 1024 characters is not stored inline. Instead:

1. The text is hashed using a content-addressed function
2. The hash is stored in the history entry’s pastedContent array
3. The actual content is stored in ~/.claude/paste-store/<hash> by hash (deduplication by content)
4. The history entry includes a reference like [Pasted text #1 +42 lines]

This saves space and deduplicates identical pastes across multiple prompts.

Paste store

The paste cache at ~/.claude/paste-store/ contains large pasted content indexed by content hash.

Structure

~/.claude/paste-store/
├── abc123def456ghi... (content hash as filename)
├── xyz789uvw012pqr...
└── ...

Each file contains the raw pasted text (no JSON wrapper). The filename is the content hash, enabling deduplication: if the same content is pasted in two prompts, only one file is stored.

References

When Claude encounters a reference like [Pasted text #2 +15 lines] in history or context, it can retrieve the content from the paste store using the content hash stored in the history entry’s pastedContent array.

Retention

Paste store entries may be cleaned up by retention policies defined in:

• cleanupPeriodDays (settings.json): How often to run cleanup
• autoCompactWindow (settings.json): Age threshold for compacting/removing old pastes

Exact retention rules are controlled by the cleanup agent; no explicit cap is enforced.

Relationship to memory

History and pastes are distinct from memory systems:

• History: Command-level index of recent prompts, not extracted findings
• Pastes: Raw pasted content, not processed memories
• Auto-memory: Extracted facts and findings from conversations, stored persistently

A single prompt may generate auto-memory entries (e.g., “Discovered that the API uses JWT tokens”) while also being indexed in history and referencing large pasted files.

---

← Back to Memory/README.md